Downhole Pressure Soft-Sensing using Interacting Multiple Modeling∗

In this work we design data-driven soft sensors of downhole pressure for gas-lift oil wells. We employ a two-step procedure. First, discrete-time (N)ARX models are identified offline from historical data. Second, recursive predictions of these multiple models are combined with current measured data (of variables other than the downhole pressure) by means of an interacting bank of (unscented) Kalman filters. We investigate the usage (i) of linear versus nonlinear models and (ii) of models with or without seabed variables in addition to platform variables. Results are validated by means of experimental data from three oil wells.